Have you seen a marathon athlete only training in a lab? Data reactive systems like LLMs need to hit the road. We will discuss real world challenges on building evaluation set for enterprise workflow systems that will instill confidence in your builds to get it on the road, just like endurance training that requires real-world exposure to terrain, weather and unpredictability. We will draw parallels from when building software changed from waterfall to agile, we needed change of tools and change of behavior. Lastly, will share some learnings from deploying LLM applications in production